Model of HSI-EfficientNetB7 for Breast Cancer Histopathology Image Analysis

Abstract

The practice of analysis is expanding in tandem with the advancement of computer science and histopathology image technologies. Combining different types of learning, such as deep learning, machine learning, and image processing, is one way to get the highest level of Precision. The purpose of the research that has been proposed is to evaluate the effectiveness of the EfficientNetB7 transfer learning approach in assessing the histology of breast cancer. This investigation is divided into three primary stages: data gathering, image categorization, and analysis. EfficientNetB7 transfer learning is the methodology that is utilized for data classification. Histopathological pictures of breast cancer specimens with a resolution of 50 x 50 were used as the source of the evaluated data (198,738 negative classes and 78,786 positive classes). Evaluation of the training accuracy, validity, and testing of breast cancer histopathological specimen images with a resolution of 50 x 50 (198,738 negative class and 78,786 positive class) obtained 91.63% accuracy (training stage) and 90.34%ccuracy (validation stage), and the accuracy result (testing stage) is 62.67%. This is the final result of evaluating the training accuracy, validity, and testing of the breast cancer histopathological specimen images. A score of 0.1158 was acquired for Cohen's Kappa, a score of 0.5422 was obtained for the F1-Score, a score of 0.6558 was obtained for Precision, and a score of 0.6267 was received for Recall for the alternative evaluation model.